機(jī)器學(xué)習(xí)技術(shù)在三維人體運(yùn)動(dòng)編輯中的應(yīng)用研究.pdf

1、廈門大學(xué)碩士學(xué)位論文機(jī)器學(xué)習(xí)技術(shù)在三維人體運(yùn)動(dòng)編輯中的應(yīng)用研究姓名：車俊飛申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)軟件與理論指導(dǎo)教師：董槐林2011ABSTRACTABSTRACTComputeranimationtechniqueshavebeenwidelyappliedforbusinesspurposesbackedbythedevelopmentofgraphicaldevicesAttheurgingofincreasingdeman

2、dforcomputeranimations，itsgenerationefficiencyisalsohighlyvaluedThus，correctandrapidcomputeranimationgenerationhasbeenaresearchhotspotinComputerGraphicsfieldRecentlymotioncapturesystemsaregettingmoreandmorepopularinprodu

3、ctionofcomputeranimations，thesesystemsarecapableofcapturingandrecordingmotiondetailsfromanactor’SbodyonspecialconditionsSOastoextendthesourceofmotiondataBasedontheaccumulationofmotiondataresearchersarepayingmoreattention

4、tomethodsandtechniquesthatcouldenhancetheeffectsandefficiencyofcomputeranimationproductionInthisdissertation，theapplicationofmotioncapturedatahasbeendiscussedfromaperspectiveofdatareusing：splittinglongmotionsequenceandbl

5、endingmotionofdifferentstylesStructureofhumanbodyiscomplex，andhasahighdegreeoffreedom，thus，basedonthehiglldimensionalityandnonlinearofhumanmotiondatamachinelearningtheoriesandtechniquesareappliedtoprocessdataofthiskindCo

6、mparedsortsofmachinelearningalgorithmsanddiscussedanewtechniqueformotiondataprocessing：(1)ManifoldlearningisemployedtocompareandanalysisthedimensionreductioneffectonmotiondataofIsometricFeatureMapping(ISOMAP)andLocallyLi

7、nearEmbedding(LLE)，classicalclusteringalgorithmKMEANSiscombinedtosplitthelongmotionsequenceExperimentalresultsshowthatISOMAPandLLEcouldcorrectlymapmotiondataofhighdimensionalityontoalowdimensionalmanifold，thus，increaseth

8、eefficiencyandaccuracyofmotionsplitting(2)Anewapproachforeditingmotionofvariousstylesiswellanalyzedusingartificialneuralnetwork(ANN)theoriesAneuralnetworkistrainedasamotionstyle，takesanothermotionsequenceasitsinputs，thus